Get started with Illumina

A hosted memory engine for AI agents. Retain raw text, recall scored context, reflect across everything a namespace knows.

curl -X POST "https://api.illumina.sh/v1/default/namespaces/demo/memories" \
  -H "Authorization: Bearer $ILLUMINA_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"items":[{"content":"Ada shipped the payments rewrite in Q3 2025."}]}'

curl -X POST "https://api.illumina.sh/v1/default/namespaces/demo/memories/recall" \
  -H "Authorization: Bearer $ILLUMINA_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"query":"who shipped payments?","budget":"mid"}'

Three core operations

Everything in Illumina builds on three operations.

A background consolidation pass groups corroborating facts into observations over time, so a namespace gets more useful the longer it lives.

Memory model

Every memory unit has one of four fact types:

TypeWhat it holds
worldObjective, external facts. The default classification.
experienceFirst-person actions and observations by the memory's owner.
decisionDurable commitments with a supersession lifecycle.
observationSynthesized by consolidation; never written directly.

Memories link to extracted entities and to each other, forming a knowledge graph you can traverse through the graph and entity endpoints. Related memories cluster into communities, and an ontology describes the entity and relationship types in play.

Jump straight in

Search docs

Search the documentation